Recognition of Material Temperature Response Using Curve Fitting and Fuzzy Neural Network

Abstract

This paper describes a system that can be used to recognize an unknown material regardless of the change of ambient tem- perature using temperature response curve fitting and fuzzy neural network(FNN). There are some problems to realize the recogni- tion system using temperature response. It requires too many memories to store the vast temperature response data and it has to be filtered to remove noise which occurs in experiment. And the temperature response is influenced by the change of ambient tempera- ture. So, this paper proposes a practical method using curve fitting to remove above problems of memories and noise. And FNN is proposed to overcome the problem caused by the change of ambient temperature. Using the FNN which is learned by temperature responses on fixed ambient temperature and known thermal conductivity, the thermal conductivity of the material can be inferred on various ambient temperature. So the material can be recognized by the thermal conductivity. Robots which can sense, think and act like man are required. Various sensors were studied to make the intelligent robot. Some contact sensors to sense force and pressure or to recog- nize forms of objects have reported, but not many a sensor to recognize material has been studied. As a fundamental study, Russell designed a sensor to recog- nize materials by thermal conductivity(1), and suggested a possibility to discriminate objects using heat conducting rela- tion. It is hard to make this method to practical use, because it takes a lot of time to reach the steady state and the characteris- tic of heat conduction is changed according to ambient tem- perature. A practical method was studied to discriminate material comparing the three points of temperature response for an unknown material with those of the look-up table in me m- ory(2). But this method has a drawback that the values are influenced by the experimental noise on the temperature re- sponses. In this paper, we propose a method in order to overcome the above problems using curve fitting of temperature response and fuzzy neural network(FNN) learned for various ambient temperatures as shown in Fig. 1. The initial transient state of temperature response(Ts) has the trend of exponential function. The exponential function approximated by curve fitting has two parameters: coefficient(C) and exponent(E). By using these two parameters, full temperature response data can be represented without noise and reserved memory. Two parame- ters were measured for the change of ambient temperature(Ta) with the interval of 5(°C). The FNN is learned by three input variables - coefficient, exponent and ambient temperature -